Jump to content

User talk:Joshfisher

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Hello, Dr. Fisher.

The tradition at Wikipedia is to add a "welcome" as the first edit to a talk page. This seems a bit silly in this case, since you have been editing much longer than I have, but welcome anyway.

I made an un-cited modification to the article on Explicitly parallel instruction computing awhile ago, and you reverted it, quite properly. I then researched the first occurrence of the name in the literature and added that, with a cite. Please check my work, if you will.

During this research, I noticed that you are a true expert and a participant in the history of EPIC. By contrast, I am an outsider, and my only knowledge is what I dimly remember reading in the trade press.

Here at Wikipedia, being an expert is a two-edged sword. On the plus side, you can tell when something is incorrect by inspection. On the minus side, you are likely to become very frustrated by the "verifiability" requirement. It took me awhile to really understand this requirement. It stems from the following:

  • Anyone can edit wikipedia
  • Anyone can claim to be anyone else: we do not have any ID verification process.
  • even if we did verify ID, we have not formal method to hold editors accountable.

Therefore, we require "verifiability" for any assertion of fact. "Verifiability" is defined as a citation to a source that can (in theory, at least) be held accountable for the assertion. Wikipedia is useful to the extent that (a) this rule is followed, and (b) "facts" reported by accountable sources are correct.

I spent a lot of time working on the Itanium article, replacing a set of three articles that were basically out-of-date press releases. If you have the time, I would appreciate it if you could review the article. Most of the references are currently to the trade press or to industry white papers, and I would like to add academic citations. Also, it is quite possible that I have over-reacted to the extreme bias of the older articles in favor of Itanium: if so, pleas give me some citable references to re-balance the article. I did try very hard to be objective.

Thanks for your time. -Arch dude (talk) 01:03, 22 May 2008 (UTC)[reply]


Itanium Stuff

[edit]

Hola Sr. Dude,

Thanks for the above message (and the kind welcome, which is even more years late than you know, since I edited a lot anonymously for a long time before I started identifying myself).

I am truly in awe of your work on the Itanium article, and incredibly thankful for all the high-quality work you have done on the Wikipedia. I was delighted to see this talk entry back when you wrote it, but have been racing around the past few weeks and responding didn't pop to the top of my queue until today. So now you get a long rambling message...

I'm well aware of the difficulty of being an expert with respect to the Wikipedia. Worse still, how does one ponder the question of things where the expert is also the only source, and we have only his memory to refer to? For example, what about the definition of VLIW? I mean, I invented the term. So does it mean what I said it meant when I invented it? What if I didn't say in so many words? Does it mean what I meant (not said) when I invented it? Does it mean what I say now I meant then? Does it mean what I say today it should mean? Well, you get the picture. It could hardly mean what people generally say it means, because there is a wide variety of uses, many of them surely wrong. In this case, I have said clearly in my book (referenced in the VLIW article--more on that later) what I say it means, and it is always what I meant, however inarticulate I was about it.

Anyway, the expert problem is pretty tricky, as are so many things on the Wikipedia. That wouldn't be the case if the Wikipedia weren't so great!

As far as your terrific Itanium article is concerned, I read through it, and as I said really admire the work you did. I know a terrifying amount about the history of the project from 1990-1994, a lot of which would probably get me into a bunch of trouble if I talked about it, despite its being many years ago. But as far as technical details--really I very much on purpose dropped out of the project the day the HP/Intel agreement was formed. I did this because the architecture style embodied in that project ran counter to my own interests, which are in much lighter-weight VLIWs, designed to be the best possible compiler targets. I admired the skill that went into Itanium, but my own inclination was to remove stuff, and that was just beating my head against a wall. I had a significant role in the project, and I'm proud of it, but it had more to do with convincing people that real compiling needed a bigger development role, and enabling that to happen. The architectural design itself wasn't my style.

So when the collaboration was announced, I put an IP wall around myself and most of my group, and moved to embedded where I thought there was much more opportunity for the stuff I liked. I did have a person working for me who worked on Itanium, but I didn't allow any technical details to pollute me or the rest of my group.

Anyway, our embedded work evolved into the ST200 family, which is a VLIW already being built into large numbers of products, and will probably be very high on the list of most popular architecture families before its time is done.

One inaccuracy in the article is that when you fixed the name history, you left in that HP called it EPIC. It did have internal names besides PA Wide-Word at HP, but not EPIC. EPIC was an Intel invention. I'll fix the article soon, or you can beat me to it. PA-WW is sufficient as an HP name, since I think that was all that was made public.

I have a lot of strong opinions on the question of whether that whole batch of architectures is or isn't VLIW. You said it nicely enough in your article. I hope you've had a chance to look at the book I mentioned above, on VLIWs and Embedded Processing. Here, for review purposes, is a call-out my coauthors and I wrote on the subject:

Controversy: Is EPIC Another Name for VLIW?

Intel took a large step in the direction of VLIW with the announcement of the Intel IPF architecture, developed jointly with Hewlett-Packard. Intel coined the term EPIC (explicitly parallel instruction computing) for the design style of this architecture, describing EPIC as building on VLIW but embodying enough new principles to deserve its own name as an architectural style. This has given rise to a small debate within the computer architecture community: Is EPIC VLIW? What separates the two? Was it more marketing considerations than technical that prompted the new name? Was Intel worried about the image of VLIW after the commercial failure of early VLIWs? Was this simply an instance of the “not invented here” syndrome? In short, do the differences between EPIC and the VLIW styles merit a new label?


There is no question that in designing this processor family HP and Intel added many features not found in previous commercial VLIWs. IPF goes farther than any prior VLIW in separating the ISA from the implementation. It exposes information so as to leave the hardware free to implement in its own way (for example, in extensive use of hints—in branch prediction, in memory latency registers, and in optional register windows). (All of these differences between EPIC and VLIW are largely there for compatibility reasons, and thus are somewhat less relevant in the embedded world.) Ironically, those searching for a distinction between EPIC and VLIW often assume it is speculation and predication that set EPIC apart—ironically because these were the hallmarks of Multiflow and Cydrome (the classic VLIW startups in the 1980s), respectively.


Ultimately, only the test of time will answer this question. Design styles are essentially art, and any real processor or architectural family will vary from the orthodoxy of any given design style (see “Fallacy: VLIW Controls Everything in Software” previously), just as any interesting painting or building will vary from the standard of a style. Whether a new variant deserves a name of its own is a subject hotly debated in many areas, as when linguists sometimes quip, “A language is a dialect plus an army plus a navy.” (Given Intel’s strength in the general-purpose processor market, this has some relevance.) In the art world, where movements with names seem to pop up daily, this subject can be most controversial. In that context and in ours, Janson’s classic History of Art has apt advice: “It has always been easier to invent new labels than to create a movement in art that truly deserves a new name.”

- Josh joshfisher 02:43, 15 June 2008 (UTC)[reply]

Thanks for the reply! I will fix the EPIC history in the Itanium article. If you cna think of any published material that sheds light on the pre-1994 history, perhaps we can expand that part of the article. -Arch dude (talk) 03:09, 15 June 2008 (UTC)[reply]

Adage

[edit]

Thanks for removing my mistaken addition about Adage in the Multiflow article. It would be even more valuable if the information could be corrected rather than removed. I don't know enough to do that myself, but perhaps something like this:

Adage, Inc. and Multiflow announced that they would merge in XXX of 19XX, but before the merger was completed, Multiflow/Adage had filed for bankruptcy.

or

Adage, Inc. and Multiflow announced that they would merge in XXX of 19XX, but when XXX withdrew its funding for YYY, the merger plans were cancelled.

or simply

Adage, Inc. and Multiflow announced that they would merge in XXX of 19XX, but the merger was never consummated.

As I say, I don't know enough to correct this passage myself; could you?

PS Jim Stockwell, one of the founders of Adage, was the father of a classmate of mine, and showed me around the Adage offices in Brighton (?) sometime in the 1960's.... Vector graphics hardware was very cool then. --Macrakis (talk) 19:31, 27 January 2018 (UTC)[reply]